University of Huddersfield Repository

Jafari, Mehdi, Shahabi, Amir, Wang, Jing, Qin, Yongrui, Tao, Xiaohui and Gheisari, Mehdi

Automatic Text Summarization Using Fuzzy Inference

Original Citation

Jafari, Mehdi, Shahabi, Amir, Wang, Jing, Qin, Yongrui, Tao, Xiaohui and Gheisari, Mehdi (2016)
Automatic Text Summarization Using Fuzzy Inference. In: Proceedings 22nd International
Conference on Automation and Computing. IEEE. ISBN 9781862181328

This version is available at http://eprints.hud.ac.uk/id/eprint/29082/

The University Repository is a digital collection of the research output of the

University, available on Open Access. Copyright and Moral Rights for the items
on this site are retained by the individual author and/or other copyright owners.
Users may access full items free of charge; copies of full text items generally
can be reproduced, displayed or performed and given to third parties in any
format or medium for personal research or study, educational or not-for-profit
purposes without prior permission or charge, provided:

• The authors, title and full bibliographic details is credited in any copy;
• A hyperlink and/or URL is included for the original metadata page; and
• The content is not changed in any way.

For more information, including our policy and submission procedure, please
contact the Repository Team at: E.mailbox@hud.ac.uk.

http://eprints.hud.ac.uk/
 Automatic Text Summarization Using
Fuzzy Inference
Mehdi Jafari
Eyvanakey Higher Institute Education
Jafari.teach@gmail.com
Jing Wang, Yongrui Qin
School of Computing and Engineering
University of Huddersfield, United Kingdom
{j.wang2, y.qin2}@hud.ac.uk
Mehdi Gheisari
Young Researchers and Elite club, Parand Branch, Islamic
Azad University, Parand, Iran
Mehdi.gheisari61@gmail.com
Abstract—Due to the high volume of information and
electronic documents on the Web, it is almost impossible for a
human to study, research and analyze this volume of text.
Summarizing the main idea and the major concept of the context
enables the humans to read the summary of a large volume of
text quickly and decide whether to further dig into details. Most
of the existing summarization approaches have applied
probability and statistics based techniques. But these approaches
cannot achieve high accuracy. We observe that attention to the
concept and the meaning of the context could greatly improve
summarization accuracy, and due to the uncertainty that exists in
the summarization methods, we simulate human like methods by
integrating fuzzy logic with traditional statistical approaches in
this study. The results of this study indicate that our approach
can deal with uncertainty and achieve better results when
compared with existing methods.
Keywords— text summarization; fuzzy logic; sense relation
I. INTRODUCTION
The summarization of a context is the process of
recognizing and indicating the most important components of a
document or a set of documents. These identified components
can be associated in a much smaller volume in comparison
with the original context. These components should have a
high accordance and relationship with the subject or concept of
that is detailed in the document [1].
Considering the scenario described above, it is very useful
for us to have an automatic system that enables us to
summarize texts and helps in the analysis of large sets of
documents [2,3]. Generally, summarization methods can be
divided into two categories: the extractive and the abstractive
summarization.
Amir Shahab Shahabi
Department of Computer Engineering, Science and
Research Branch, Islamic Azad University, Tehran, Iran.
shahabi_amir@azad.ac.ir
Xiaohui Tao
Faculty of Health, Engineering and Sciences
University of Southern Queensland, Australia
xtao@usq.edu.au
In the extractive method, some properties for each part of
the text are determined; and based on these properties and
characteristics, the level of importance of each part of the text
is determined and finally best of them will be selected as a
component of the final summary. But in the abstractive
methods, usually the techniques of the natural language
processing are used. And because of difficulties in general
natural language processing, it is usually difficult to achieve
satisfied results. The abstractive methods are usually a
combination of both extractive and abstractive methods [4].
Because of a lack of certainty in the text summarization and to
deal with uncertainty, fuzzy logic [5] is used. Fuzzy logic is
used to measure the degree of importance and correlation and
also to highlight the important phrases to create summarization.
In this work, we integrate fuzzy logic with traditional
extractive and abstractive approaches for text summarization.
Through such technique, we obtain a summary of the original
text, which can best reflect the main delivery of the original
context.
The remainder of the paper is organized as follows. In
Section II, we review related work on text summarization.
Then our approach is detailed in Section III. In Section IV,
experimental evaluations are presented to shown the benefits
our approaches over exiting approaches. Finally, we conclude
our work in Section V.
II. RELATED WORK
In [6], [7], methods for text-based summarization by using
the fuzzy logic and fuzzy inference system are presented.
Genetic Algorithm and Genetic Programming are also used to
optimize the rule sets and membership functions of the fuzzy
system.
 The results of the comparison of the fuzzy-based methods
and other methods have been achieved [8]. The main problem
of previous work is that only syntactic parameters and semantic
parameters are used. But the semantic relationship between
words is ignored. Ignorance of the words semantic relations
and sentences and the sole attention to the syntactic structures
of the words and the sentences is mainly because that they
could not help improve the accuracy of text summarization. On
the other hand, if we just pay attention to semantic relations of
expressions but ignore syntactic structures, it does not improve
the accuracy of text summarization either [9]. In this paper, we
propose applying both semantic relations and syntactic
structures of expressions in producing text summarization and
using the combination of these two aspects to improve the
quality of summarization.
III. FUZZY INFERENCE BASED TEXT SUMMARIZATION
The overall flowchart of text summarizer proposed in this
paper is given in Figure 1. In the preprocessing phase, to
convert the original text into the text that will be used as an
input parameter in the summarization system, the following
steps should be taken:
• Remove redundant words. Ordinary words that do
not have any specific data and do not show any value
will be deleted, such as "the", "an", "a", etc.
• Case folding. Either uppercase letters are converted
into lowercase or all lowercase are converted into
uppercase. Here we convert all characters into the
lowercase.
• Stemming. The derived words are converted into
their stemming. For example, the names that are in
the plural forms should be converted into their single
forms and the verbs should be converted into their
original forms. It should be noted that here we do not
include the stemming phase to extract the semantic
parameters, because we are using databases such as
Wordnet [10] and we want the main words to be
protected.
After the preprocessing phase is finished, the parameter
extraction phase will start. In the following, syntactic and
semantic parameters that are used in this paper are described.
A. Syntatic Parameters
TF/ISF (Term Frequency–Inverse Sentence Frequency):
This parameter is actually inspired by the information retrieval
fields, where the TF/IDF (Term Frequency–Inverse Document
Frequency) concept is introduced. Generally the parameter
TF/IDF is used in multiple documents summarization but in
this paper, it is used for a single document that is converted
into TF/ISF. To calculate this parameter, first the parameter TF
is calculated for each word. The value of the TF of each word
equals to the number of occurrences of the word in a document,
divided by the total number of the words. Then we get the
value of ISF for each word that is equal to the number of
statements that include that word, e.g., Nterm dividing the total
number of the sentences N. Thus the value of TF/ISF for each
word is given in Equation (1):
TF/ISFterm = TFterm *lg (Nterm/N) = TFterm * ISFterm (1)
Then, to compute the TF/ISF values of each statement, the
values of TF/ISF for all words are added up together and we
obtain the TF/ISF values of a sentence. We normalize the
TF/ISF by dividing this using the maximum value of TF/ISF
among all the sentences.
Preprocessing
phase
Summary
Production
Extraction of
the syntactic
parameters
Calculating of
sentence grades
based on fuzzy
Extraction of logic
the semantic
parameters
Figure 1: Flowchart summarizing the proposed approach
Sentence Length: This parameter is in fact the total
number of main words. But very short sentences will be
removed by the penalty and then they will not be placed in the
summary. Dividing the largest sentence can help to normalize
other parameters.
Location of Sentence: We can consider the relative
position of a sentence in the whole document or in the section
or in the paragraph, and so on. Here, the entire document will
be considered in Cartesian Space and each sentence has an “x”
and a “y” value.
The value of “y” is the number of paragraph that sentence
is placed and the value of “x” is the number of sentence in that
paragraph. Only five sentences of each paragraph are
considered important. The location value of a sentence in the
paragraph is the inverse order of the sentence number i.e. the
first sentence has the value of the 5/5 and the second sentence
has the value of 4/5, and so on [11].
Similarity to the Title: We use the vector method to split
the title words and separate the words of the sentence. Then we
use the cosine measure [12] to measure similarity between
sentences and the title of the document. More details about
how to convert sentences and titles into vectors should be
referred to [12]. Then similarity values will be calculated as in
Equation (2) (according to reference [12]):
.
!"!#$%$&! !"#$"%&
SimToTitlesen= (2)
!"!#$%$&! ∗ !"#$"%&
 Similarity to Keywords: This parameter is the same as the
previous parameters and to calculate it, the cosine measure
between the desired vector and the vector of keywords can be
calculated. The ten words that have the highest TF/ISF
throughout the document are considered keywords.
Text-to-text coherence: For each sentence, we add all
values of that sentence with other sentences in the same
document. In order to calculate the similarity value between
two sentences, we again use the cosine measure similar to the
way of computing Similarity to the Title and Similarity to
Keywords. Then we normalize this value by dividing it using
the obtained maximum value for all sentence similarities.
Integrated text-to-center: Initially, we get the central
sentence by using the sentence location parameter as described
above. Then, for each sentence, the similarity value of that
sentence is obtained and normalized by dividing it using the
maximum obtained value among all sentences.
Key Concepts: The main concepts actually are 15 distinct
words with the highest TF/ISF that have the following
property: all words should be nouns. If a sentence contains
those words it should be included in the final summary,
otherwise not. But this does not consider sentences that contain
only some of the 15 key words. For example, a sentence may
contain a word of the main concepts and another sentence may
contain 5 words of the concept words. To distinguish all these
cases, we use fuzzy logic.
The nouns: Specific names/nouns may refer to people,
places, and so on. These nouns are critical in generating a
better summary. If a sentence contains specific names/nouns, it
should be included in the final summary, otherwise not.
However, we need to be selective during this process. If a
sentence contains more nouns, it would tend to be a better
sentence. Here, we apply fuzzy logic again to determine the
selection of the final sentences. To identify the specific nouns,
if the word is also not in the Wordnet database, it will be
considered as a specific noun.
Non-basic information: If expressions (speech markers),
such as because, while, in addition, and so forth, which usually
occur at the beginning of the sentence, are in a sentence, we
say that sentence contains non-basic information. If any
sentence includes such expressions, the sentence is likely to be
the emphasizing sentences, explaining sentences, or proverbial
sentences. Such sentences are that we should avoid to include
in the summary.
Anaphors: This parameter is the repetition of an expression
in the successive sentences. The anaphors actually contain non-
basic information and if a sentence contains an anaphor, its
contents are covered by other related sentences. To compute
this parameter, we will examine each sentence by its adjacent
sentences. If up to six words are recurring, then the examined
sentence is an anaphor.
After obtaining the syntactic parameters, then we need to
go through the semantic parameters that are described below.
B. Semantic Parameters
The linguistic parameters use semantic knowledge such as
semantic relationships between words and their combinational
syntax. Based on such information, we can determine which
terms are similar to each other.
Semantic similarities between sentences: Semantic
similarity of two sentences is well discussed in [13]. To
calculate this parameter, we first calculate the vector value
properties for each sentence, then the maximum value of the
semantic similarity between the words in the vector properties.
The words in the same sentence are used to determine the
words weight. Only the word similarity is used as the words
weight. Further, the two words that are in the set of a part-of-
speech are compared. Here, to obtain the semantic similarity
between sentences, we use the approach that is presented in
[14]. Similarity of two sentences is calculated according to
Equation (3).
Σ!!!"! !"# !!!"! !"#$%&(!!,!!)
Simsemantic(S1,S2)= (3)
!! ! !!
The value of semantic similarity of each word in the first
sentence with the all words that belong to the same part-of-
speech in the second sentence is calculated and we retain the
words that achieve maximum similarity. Then the summation
of all maximum similarities is calculated and divided by the
summation of the lengths of the two sentences and finally gets
normalized.
The process of the calculation of semantic similarity
between two words is as follows. Here, similar to [15],
semantic similarity between two words is computing using
Equation (4) in our approach.
!∗!!
WordSim(x1,x2)= (4)
!!!!!!!∗!!
Where N1 is the number of the links with the father of
sentence x1 until receiving the first common father with
sentence x2. And N2 is also the number of links of x2’s fathers
until receiving the first common father of x1 and N3 is also the
number of links with the common father to the root. For more
details, interested readers are referred to [15].
The Order of Words: The composition of the sentence
words plays an important role in the concepts of the sentence.
This results in that the order of the words in the sentences is
also important. For example, consider the sentences {the sale
manager hits the office worker} and {the office manager
hits the sale worker}. Indeed, these two sentences contain the
same set of words but different sequences of words lead to
different meanings. Recognition of the sequence of the words
in a sentence is easy to the human but it is very challenging in
natural language processing. Here, we use the method that has
been proposed in [16]. In order to calculate the similarity of
sequence of the words between the 2 sentences, at first we
create the Join Set through the 2 sentences. First the words of
the shorter sentence and then the words of the larger one are
inserted into the Join Set. For example, the join set of the
previous two sentences is {the sale manager hits the office
worker}. Then the sequence vector of each of the two
sentences is calculated. For example, for each word in the set
of Join Set, if the same word in the same sentences exists in the
corresponding place of that word in the sequence vector, the
evidence index of that word in the sentence will be used. If that
word does not exist, the index of the similar word of that word One of the problems of conventional logic is in the
that is larger than the threshold value will be used. For all other constraints of both values: True or False. However,
cases, the value of that index is 0. conventional logic is well suited for a two-state system, but it is
not suitable for systems that are not certain. The most
After obtaining the sequence vectors of the two sentences important disadvantage of the conventional logic in a
r1 and r2, by using Equation (5) the difference value between
summarization process is that some parameters that are
the two vectors can be calculated. mentioned above are not two-state. As an example, to say that
SimwordOrder(S1,S2)=1-
!!!!!
(5) two sentences are semantically similar, or not, is not correct but
!!!!! in fuzzy logic, we can say that how much the two sentences are
The value is also normalized. One special case is that the similar to each other. In our work, fuzzy logic is used for
word that is in the Join Set does not exist in the corresponding measuring the degree of importance and correlation and also
indices. But by using corpus-based knowledge bases such as used to identify the important sentence to create
Wordnet the fact is that, we can use the method proposed in summarization.
[17] to achieve the computation of similarity on top of word
IV. EVALUATIONS
orders. In [17], the LCS1 is used to determine the similarity
between two words. A method for normalization is dividing the In this paper, we verify our approach using a real-world
length of LCS into the longer string but because small strings dataset. In proceedings of JAIR, 50 randomly articles was
are not used, it is not a suitable solution. This is because in selected. The average number of words in an article is 436
some cases the small strings are very important. Therefore, we words. First, all tables, images and formulas are eliminated in
use Equation (6) for computing the normalization. the preprocessing phase. The evaluation of summarization, in
both internal and external evaluation, is carried out. In internal
!"#$!!(!"#(!,!))!
NLCS(r,s)= (6) evaluation the focus is on the quality of the summarization,
!"#$!! ! ∗!"#$!!(!)
while in the external evaluation, most of focus is on the
performance of the system in a particular problem. In fact,
The LCS method is case-sensitive but because of there is not consensus on which method is better or is
converting all letters into lowercase in the preprocessing phase, preferred. But since the most recent work has been on internal
the computation is ensured to be valid. evaluation and because of requirement of fair comparisons, in
C. Combination of Parameters this paper we use the internal evaluation.
In this paper, after calculating the semantic similarity The quality of evaluation in this paper that is summarizing
between the sentences and the similarity of word sequences in by the machine is compared with the summarizing of the
the sentences, the overall semantic similarity of the two proceeding JAIR issues that is summarized by human. Our
sentences can be obtained via Equation (7). assessment is also based on the F2 scale, which uses the values
TotalSimsemantic(S1,S2)=a*Simsemantic(S1,S2)+(1-a) SimWordOrder (S1,S2) (7) of scales P (Precision) and R (Recall). The general formula for
P, R and also F that we use, are shown in Equations (8) to (10).
In [18] and [19] it is proven that the combination of
parameters is greater than the syntax parameter, and thus is the (!"#$%& !" !"##$!% !"#$"#%"! !!!" !"#$%&#!' !" !"##$%&'(% !"!#$% )
best evaluation of sentence similarity. The parameter a in our P= (!"#$% !"#$%& !" !"#$"#%"! !!!" !"#$%&#!' !" !"##$%&'(% !"!#$%)
(8)
experiments takes 0.8 in the above equation.
D. Fuzzy Logic P=
(!"#$%& !" !"##$!% !"#$"#%"! !!!" !"#$%&#!' !" !"##$%&'(% !"!#$% )
(9)
(!"#$% !"#$%& !" !"#$"#%"! !!!" !"#$%&#!' !" !"##$%&'(% !"!#$%)
A lot of work has been done on summarization based on
syntax parameters. Also the several semantic parameters are
!∗!∗!
discussed in several papers. Also in [20] [7] [6], it has been F= (10)
!!!
shown that the use of fuzzy logic achieves positive results in
improving the summarization results. But our goal in this paper
Here, P measures the accuracy of the measure (Precision) and
is to combine these cases and according to our knowledge,
R is Recall.
there is not any similar work in this area. The results also show
that combining both syntactic and semantic parameters, and
Table 1 Comparing the results of summarizing methods
using the fuzzy logic can improve the results. That indicates the
quality of our hybrid approach.
MS Word Copernic Presented Method
As classic logic is the base of the logic-based expert
Avg Avg Avg Avg Avg Avg Avg Avg Avg
systems, fuzzy logic is also the base of the fuzzy expert R P F R P F R P F
systems. In fact, in facing the uncertainty when we use the JAIR 0.28 0.31 0.29 0.46 0.55 0.51 0.58 0.6 0.59
fuzzy logic, modeling processes that use such techniques are
more favored compared with conventional logic.

1 2
Longest Common Subsequence Fitness
 Table 2: Comparison of proposed method with quality Huang
Huang Method Presented Method
Avg F Avg F
JAIR 0.46 0.58
Initially, measure F is calculated separately for each
summarized essay then the average of F is obtained for all sets
of the essay. The results of the summarizing method presented
in this paper are shown in the Table 1. As can be seen the
presented summarizing method shows better quality in
comparison with other summarizing methods.
Moreover, comparisons with the mechanical summarizer
that is presented in [21] are also performed on this set of
essays. The comparison is shown in Table 2. We call the
proposed method in [21] in Table 2 as the Huang Method. As
can be seen in Table 2 the proposed summarizing method
works better than the existing method, and shows better
quality, although since the summarization process is not certain
and is related to individual interest of the linguistic experts, the
summarizing quality is not close to 100% correctness.
Overall, the proposed method shows better quality in
comparison with existing summarizing methods.
V. CONCLUSIONS
The perception of a context from a computational
perspective is still an unsolved problem. Many existing
summarizing methods are mainly based on statistical and
probabilistic extractions. These methods have no perception
from the concept and meanings of the context. Summarizing
most of the texts needs deeper understanding of the concepts to
be meaningful.
We observe that attention to the concept and the meaning of
the context could greatly improve summarization accuracy, and
due to the uncertainty that exists in the summarization
methods, we simulate human like methods by integrating fuzzy
logic with traditional statistical approaches in this study. The
results of this study indicate that our approach can deal with
uncertainty and achieve better results when compared with
existing methods. Our proposed method has been demonstrated
to be able to provide better summarizing quality in comparison
with existing methods.
Acknowledgement
The authors wish to thanks from deputy of research of Islamic
Azad University Science and Research Branch for its support
of this project and Dr Mohammad Mehdi Esnaashari from
ITRC.
REFERENCES
[1] Mani, Inderjeet. Automatic summarization. Vol. 3. John Benjamins
Publishing Company, 2001.
[2] E. Liddy, “Advances in automatic text summarization,” Information
Retrieval, vol. 4, no. 1, pp. 82–83, 2001.
[3] P. P. Balage Filho, T. A. Salgueiro Pardo, and M. das Gracas Volpe
Nunes, “Summarizing Scientific Texts: Experiments with Extractive
Summarizers,” in Intelligent Systems Design and Applications, 2007. ISDA
2007. Seventh International Conference on, 2007, pp. 520–524.
[4] J. Steinberger and K. Ježek, “Update summarization based on latent
semantic analysis,” in Text, Speech and Dialogue, 2009, pp. 77–84.
[5] L. A. Zadeh, “From circuit theory to system theory,” Proceedings of the
IRE, vol. 50, no. 5, pp. 856–865, 1962.
[6] F. Kyoomarsi, H. Khosravi, E. Eslami, P. K. Dehkordy, and A. Tajoddin,
“Optimizing Text Summarization Based on Fuzzy Logic,” Seventh
IEEE/ACIS International Conference on Computer and Information Science
(icis 2008), pp. 347–352, May 2008.
[7] a. Kiani-B and M. R. Akbarzadeh-T, “Automatic Text Summarization
Using Hybrid Fuzzy GA-GP,” 2006 IEEE International Conference on Fuzzy
Systems, pp. 977–983, 2006.
[8] F. Kyoomarsi, H. Khosravi, E. Eslami, and P. Khosravyan, “Optimizing
Machine Learning Approach Based on Fuzzy Logic in Text Summarization,”
International Journal of Hybrid Information Technology, vol. 2, 2009.
[9] P. Achananuparp, X. Hu, and X. Shen, “The evaluation of sentence
similarity measures,” Data Warehousing and Knowledge Discovery, pp. 305–
316, 2008.
[10] M. M. Stark and R. F. Riesenfeld, “Wordnet: An electronic lexical
database,” in Proceedings of 11th Eurographics Workshop on Rendering,
1998.
[11] L. Suanmali, M. S. Binwahlan, and N. Salim, “Sentence Features Fusion
for Text Summarization Using Fuzzy Logic,” 2009 Ninth International
Conference on Hybrid Intelligent Systems, pp. 142–146, 2009.
[12] S. Fisher and B. Roark, “Query-focused summarization by supervised
sentence ranking and skewed word distributions,” in Proceedings of the
Document Understanding Conference, DUC-2006, New York, USA, 2006.
[13] Mojtaba Sedigh Fazli, Jean-Fabrice Lebraty. A solution for forecasting
pet chips prices for both short-term and long-term price forcasting, using
genetic programming. WorldComp’2013. The 2013 International Conference
on Artificial Intelligence, Jul 2013, Las Vegas, Nevada, United States.
CSREA Press, II, pp.631-637. <hal-00859457>.
[14] R. Malik, L. V Subramaniam, and S. Kaushik, “Automatically selecting
answer templates to respond to customer emails,” in Proceedings of the 20th
international joint conference on Artifical intelligence, 2007, pp. 1659–1664.
[15] Mojtaba Sedigh Fazli, Jean-Fabrice Lebraty. A comparative study on
forecasting polyester chips prices for 15 days, using different hybrid
intelligent systems. IEEE & International Neural Network
Society. International Joint Conference on Neural Networks, Aug 2013,
Dallas, Texas, United States. pp.1869-1875, 2013. <hal-00859445>
[16] Manaf Sharifzadeh, Saeed Aragy, Kaveh Bashash, Shahram Bashokian,
and Mehdi Gheisari. A Comparison with two semantic sensor data storages in
total data transmission. SoftConf, March 2013, Sanfransico
[17] A. Islam and D. Inkpen, “Semantic text similarity using corpus-based
word similarity and string similarity,” ACM Transactions on Knowledge
Discovery from Data, vol. 2, no. 2, pp. 1–25, Jul. 2008.
[18] T. K. Landauer, D. Laham, B. Rehder, and M. E. Schreiner, “How well
can passage meaning be derived without using word order? A comparison of
Latent Semantic Analysis and humans,” in Proceedings of the 19th annual
meeting of the Cognitive Science Society, 1997, pp. 412–417.
[19] P. Achananuparp, X. Hu, X. Zhou, and X. Zhang, “Utilizing sentence
similarity and question type similarity to response to similar questions in
knowledge-sharing community,” in 17th international conference on World
Wide Web, 2008.
[20]Mehdi Gheisari, Ali Akbar Movassagh, Yongrui Qin, Jianming Yong,
Xiaohui Tao, Ji Zhang, Haifeng Shen, "NSSSD: A New Semantic
Hierarchical Storage for Sensor Data", IEEE 20th International Conference on
Computer Supported Cooperative Work in Design (CSCWD 2016), May 4-6,
2016, Nanchang, China
[21] Huang, Hsun-Hui, Yau-Hwang Kuo, and Horng-Chang Yang. "Fuzzy-
rough set aided sentence extraction summarization." Innovative Computing,
Information and Control, 2006. ICICIC'06. First International Conference on.
Vol. 1. IEEE, 2006.
[22] Porter, Martin. "The Porter stemming algorithm, 2005." See http://www.
tartarus. org/~ martin/PorterStemmer.
[23] M.Gheisari,” Design, Implementation, and Evaluation of SemHD: A
New Semantic Hierarchical Sensor Data Storage”, Indian J. Innovations Dev.,
Vol. 1, No. 3 (Mar 2012) ISSN 2277 – 5390